Recording evidence an inspector will accept

An evidence record answers a question asked years later by somebody who was not there. That is a higher bar than it sounds, and most records fail it by omitting the things that were obvious to everybody in the room at the time.

Treat failure as a first-class outcome

The most common mistake is designing the happy path and treating a failure as an exception to be handled elsewhere. In practice this means failures end up in an email thread, and the register shows an unbroken run of passes.

An unbroken run of passes is not reassuring to an experienced auditor. It reads as a register that is not being used honestly, and it invites exactly the digging you did not want. Record passed, failed and conditional as equal outcomes, on the same form, in the same table.

As-found and as-left

For anything measured — a calibration, a pressure test, a containment check — record the value before adjustment and the value after. The as-left value on its own tells you the equipment is fine now. The pair tells you how far out it had drifted, which is the thing that actually informs the next interval.

It also answers a question you cannot answer retrospectively: if the instrument was out of tolerance when found, what was produced or tested with it since the last check? Without the as-found value, that investigation has no starting point.

• The measured values, with units and the tolerance they were judged against.
• The instrument or standard used, and its own calibration status at the time.
• The person who performed it, and the person who reviewed it if those differ.
• The date, and the certificate or report as an attached document.

Link the correction to the finding

When a result triggers a corrective action, the two records should be one hop apart. Not in a separate system, not connected by a reference number somebody types into a notes field — an actual link, so that opening the failure shows what was done about it.

The chain an inspector wants to walk is short: finding, action, verification, closed. If assembling that chain takes a person and an afternoon, the evidence exists but the system is not doing its job.

Never let a failure be quietly replaced by the passing re-test that followed it. The re-test is a second record, not an edit of the first.

Write for the stranger

The last rule is editorial. Write the result so it makes sense to somebody with no memory of the day: expand the abbreviation, name the machine rather than saying the usual one, and say what conditional actually meant here. It costs thirty seconds now and it is unrecoverable later.